AI maturity is capped by data maturity

Almost every PE-backed platform we talk to has already bought some AI. Usually it’s a ChatGPT Enterprise rollout, or a copilot bolted onto the ERP, or a vendor demo that looked incredible in the room. Very few of them can point to a single number on the P&L that moved because of it.

The problem is rarely the tool. It’s the order things got done in.

If your data lives in five systems that can’t agree on what the word “revenue” means, no model is going to reason across your portfolio for you. The ceiling on what AI can do for you was set long before you picked a vendor. It was set by the state of the data underneath, and you can’t spend your way past a foundation that hasn’t been built yet.

So the more useful question isn’t which AI tool to buy. It’s where you actually stand today, and what the next realistic rung up looks like. We tend to map that out as a ladder.

The ladder

There are roughly six levels to it. At the bottom is the AI your staff are already using on their phones without telling you. At the top is an operating model where AI is the value-creation thesis itself rather than a feature you bolted on later. Most platforms are stuck somewhere in the bottom third, and the thing that traps them is simple: you can’t climb by spending more on the rung you’re already standing on.

Here’s how it stacks up, top to bottom.

The trap gets obvious once you can see the shape of it. The bottom two rungs, shadow AI and licensed copilots, are where nearly everyone lives. They’re real and they save people time, but they don’t show up in the financials. The serious operating leverage doesn’t appear until Tier 4, where agents start doing actual work: drafting the quote, reconciling the PO, flagging a pricing anomaly, putting together the board pack. So the natural instinct is to leap straight from Tier 1 to Tier 4, put “AI on top of the ERP,” and call that a strategy.

It falls over, because the rung in the middle got skipped.

Tier 3 is the part nobody builds

That skipped rung is cross-entity structured intelligence. One source of truth spanning every operating company, where margin and pipeline and utilization each get defined exactly once and then used everywhere. Think of it as the company brain. It’s the real starting line for doing anything useful with AI, and it’s the part most roll-ups never get to.

The reason they never get to it is structural. Every acquisition arrived on its own stack. The board got its KPIs stitched together by hand, weeks late, and the numbers still didn’t tie out. There was never one governed place where the data actually lived. Drop an agent on top of that and what you’ve built is an agent working off a partial picture. It’ll be confident, it’ll be fast, and it’ll be wrong.

A Tier 4 agent is only ever as good as the Tier 3 data it reasons over. Get Tier 3 right and the frontier work stops being a science project. Skip it and the frontier work stays a demo, forever.

The part that surprised us: two phases, at the same time

The usual way to read a maturity ladder is one rung at a time, in order. In practice that’s not how the good engagements go.

The two rungs that matter most are the ones you run side by side.

Tier 2, document and knowledge automation, doesn’t need a warehouse at all. It’s “ask the company anything,” grounded in your own spec libraries, submittals, manuals, and old bids. It pulls decades of tribal knowledge out of people’s heads before they retire, it’s cheap to stand up, and it starts paying off almost right away. You can have it live in a few weeks.

Tier 3, the warehouse and the governed layer beneath it, is the heavier build. That one runs in months, not weeks.

So we do both at once. Tier 2 banks visible wins early and buys you the credibility to keep going, while Tier 3 gets built quietly underneath. By the time the document assistant has paid for itself, the warehouse is standing, and now the frontier rungs are actually within reach. Nobody has to sit on their hands for two quarters waiting to see whether any of this works.

What this means if you’re buying companies

If you’re acquiring at any kind of speed, none of this stays abstract for long. Every deal that lands on a new system is one more reason your data maturity stays low, and your AI ceiling sits right on top of it. What fixes that isn’t a smarter model. It’s a foundation that turns each new acquisition from a six-week integration project into a day or two of onboarding, with every entity flowing into one place and every metric defined once.

That same foundation is what puts the top of the ladder back within reach. It also happens to make the platform worth more when you go to sell, because the data and agent layer becomes something every future acquisition plugs straight into instead of a cost that resets with every deal.

Build it once, own it outright, and AI quietly stops being a line item you keep having to justify and starts being part of why the thesis works at all.

---

This is the lens we bring to every engagement. Work out which rung you’re actually on, run Tier 2 and Tier 3 at the same time, and only reach for the frontier once the base can take the weight. If you’re a multi-entity operator or a PE-backed platform trying to turn AI spend into something that shows up on the P&L, that’s the work we do.